How to give your AI assistant its own emailaddress in Grok Bot (and the traps I hit)

Your AI assistant can now have its own email address, and it's free with Grok Bot. For me it works way better than the timer-checked inbox I had before, as long as you set the sender limit before a client ever sees the address.

Grok Bot started handing these addresses out on Friday, October 9. 9to5Mac reported the rollout the same day: you claim one by asking your bot.

If you've ever pasted a client's email into an AI chat and pasted the answer back, you've been the courier in your own setup. With an address, your client writes to the assistant like they'd write to you. You still approve what goes back, but you stop carrying it.

I was a courier too, for my own website assistant. It had an inbox on AgentMail, an email service for AI assistants, and I still had to remind it to check. In late September I typed, "please set this up so i'm not always pinging you to remind you."

Grok Bot is the app I run my assistants in. Here's the order I'd set this up in now, and the traps I hit along the way.

Step 1: In Grok Bot, claim the address and name it like a business email

Ask your assistant to "get yourself an email address." According to Grok Bot's email guide, it asks what name you want, checks that it's free, and shows you an approval card. Nothing is claimed until you approve.

Pick the name the way you'd pick a business email, because you're stuck with it. Each account gets one address, shared by all your assistants, and you can't rename or delete it.

On launch day I tried to claim a second one, so each of my assistants could have its own. The limit held. I haven't found a way to buy more.

Step 2: Tell it what to do when mail arrives

An address alone does nothing. The guide is plain about it: "New mail doesn't start a chat or wake a Bot on its own." Your assistant acts on incoming mail only through a routine, which is an instruction it runs when something happens.

So type something like "When an email arrives from my client, draft a reply and show me." By default, that routine only runs for senders that pass standard email authentication checks, and mail that fails the spam scan never starts it.

That arrival trigger is what my old setup was missing. It ran on a timer. In late September I'd asked my website assistant to check its inbox every 15 minutes instead of every 30. It worked at first, then stopped working over a weekend. I never found out whether the mailbox or the timer broke.

A timer also costs you something when nothing happens. Grok Bot's routines guide says each run spends usage, so a check that finds an empty inbox still counts.

My other assistant uses the Grok Bot email. A new message starts it right away, so there's no timer to stop.

Step 3: Set limits before your client gets the address

Name your client's address, or their company's domain, when you ask for the routine. The guide's own example works the same way: you name the sender in the request. A routine can watch mail from anyone, from specific addresses, or from a whole domain.

Then open the routine and check that "When to run" says what you meant. That field shows the saved trigger in words, so you can catch a routine that would wake up for the wrong people.

I'd keep that limit there, outside the prompt, where the agent can't argue with it. A sender filter on the trigger decides whether the routine runs at all. A rule in the instructions is something the agent reads and then interprets.

Some limits have to live in the instructions anyway. Routines can't filter by subject, so a rule like "website changes only" goes there.

Replies have their own guardrail. Your assistant sends email only when you've asked it to, in chat or in the routine's instructions. Otherwise it shows you the full draft and asks first. So keep "draft a reply and show me" in the instruction while you test.

Two more things before you hand the address out. Tell your client they're writing to an assistant. Then email it from an unlisted address. Nothing should run.

Step 4: For a second address, use a mailbox with an alert

Since Grok Bot gives each account one address, a second assistant needs a mailbox somewhere else. What matters is that the mailbox can send an alert the moment mail arrives, so your assistant isn't checking on a timer. These alerts are usually called webhooks, and a Grok Bot routine can start from one.

I kept AgentMail and turned one on. AgentMail's webhook docs describe it as a request it sends "as soon as an event happens," instead of you asking over and over whether there's new mail.

The next morning, my website assistant set up the routine and pointed me to a link and a key to paste into AgentMail's settings. It couldn't see those values from its side, so I copied them myself. Then it reported picking up a test email within a minute. That's its report; I didn't time it myself.

The catch: the Step 3 sender limit doesn't cover AgentMail. Grok Bot's sender filter belongs to routines started by its own email, so on this path my assistant's only filter is its instructions, which tell it to act only on mail sent to its own address.

That's the place I just said the agent can argue with. So run the same test here: email it from an unlisted address, and check what the assistant did with it.

My order now

The Grok Bot email first, and AgentMail for the rest.

In my use, AgentMail with an alert is almost as good. I've tested it once, so I can't call it reliable.

Anyway.

Your client already knows how to send an email. When I swapped out my assistant's setup that next morning, the person who emails it kept the same address and didn't have to change a thing.

What's one email you get every week that you'd let an assistant answer first?